[Talk Summary 1] Web as a textbook: Curating Targeted Learning Paths through the Heterogeneous Learning Resources on the Web

In this seminar, Igor presents an idea about organizing heterogeneous educational resources on the web into structure alike to a textbook or course. Thanks to the structure, engines might be able to allow learners to navigate a sequence of webpages that take them from their prior knowledge to material they want to learn. He gives an opinion that educational resources on the internet are diversity; they could be articles, lecture notes, tutorials, slides, etc. And those materials are provided by various kind of creators from different perspectives, and thus feed a variety of learners who do not necessarily rely on textbooks.

To approach this task, Igor first presents a document as a bag-of-technical-terms consisting of two multi sets, a set of Explained terms and a set of Assumed terms.

• Explained: the term appears in the context and is explained to be understood by readers. 
• Assumed: the term corresponding to a explained term is assumed to be familiar with readers, and is required for understanding the context in which it appears.

As an illustration, consider the following excerpt from Christopher Bishop’s classic textbook Machine Learning and Pattern Recognition from the chapter that introduces the concept of Expectation Maximization:

Expectation Maximization
An elegant and powerful method for finding maximum likelihood solutions for models with latent variables is called the expectation maximization algorithm, or EM algorithm.

 

In the excerpt above, the blue terms appear in the Explained aspect and the red terms that appear in the Assumed aspect. Understanding the concept of Maximum likelihood is a prerequisite for understanding Expectation Maximization.

After that, he uses this representation of documents to connect web sources which explain concepts to those web sources where the same concepts are assumed.  Document A could be a prerequisite of document B, if the assumed concepts in B are explained in A. As a result, the system can determine to path to document B which assure readers able to understand B.

To archive that goal, he presents:

1. A supervised classification approach to identifying explained and assumed terms in a document.

2. An algorithm for finding optimal paths through the web resources.

Based on the work of automatic curriculum extraction, Igor gives an open idea of modeling users' knowledge by a set of terms they already know and recommending learning paths to archive their learning goals.

Question and Discussion:

1. A model for estimating users' knowledge state should be very challenging. We probably accept that the knowledge of users could be presented as a number of terms they understand (or at least they know). But to determine terms that users understand in the context of studying heterogeneous educational resources on the web is significantly difficult. How do we know learners understand a concept just by their reading activity on the internet? 

2. Are there more than one definition for a concept? Because a concept is defined just by a set of other concept around it, there are probably various definition of the concept depicted by different perspectives of creators about the concept.

3. A great number of different resources about one topic potentially make noise to users if we just consider occurring explained and assumed terms  to recommend them without the reliance of resources.

09/06/2016

Room 828, IS Building, 

135 North Bellefield Avenue,

Pittsburgh